Abstract
This paper is a report on the work done by the LDOS team (from the University of Ljubljana Faculty of electrical engineering) on the 2011 RecSys CAMRA (Challenge on Context-aware Movie Recommendation). We present three approaches in the track 1 competition which requires to select the top N recommended items for each household. Our general approach uses a matrix factorization algorithm to compute per-user rating predictions. The context is taken into account using an averaged and weighted sum to calculate the household-based rating predictions. We also explored how the limiting of true positive candidates with additional knowledge from the dataset influences the performance of our models.